Wednesday, August 27, 2008

Gustav and uncertainty

I'm watching the forecasts for Tropical Storm Gustav with dread.


Image: five-day track forecast for Gustav, from NOAA.

Three years ago this week, I began my first Earth Systems Science lecture with an image of a swirling cloud looming over the Gulf of Mexico. On the second day of class, I was asking students if they knew anyone from New Orleans, and whether they were all right. It was horrifying, particularly because I had used the geology of New Orleans as an example in classes for years, ever since I read The Control of Nature.

But even though Gustav appears to be headed on the absolute worst course for the city of jazz, levees, and elevations below sea level, Forecaster Franklin of the National Hurricane Center added this reminder to the end of Wednesday's 11 pm discussion:

IT PROBABLY WOULDN'T HURT TO REMIND EVERYONE THAT THE AVERAGE 5-DAY OFFICIAL TRACK ERROR IS ABOUT 300 MILES...AND THE AVERAGE 5-DAY INTENSITY ERROR IS ABOUT 25 MPH.


Error. In other words: there are reasonable odds that we'll be wrong, and the storm will hit Houston or the Florida Panhandle.

I'm not a statistician. In fact, I've never taken a statistics class. (I'm embarrassed enough to have bought an undergrad statistics textbook, but not shamed enough to have buckled down and worked through it.) But I am a scientist and a teacher, and that means that I need to somehow get students thinking about uncertainty.

I'm not a weather modeler, so I think about uncertainties more in the context of measurements than of predictions. I learned about the concepts of accuracy and precision in chemistry classes, and about error estimates in intro physics. My geology classes never took on the topic directly. But it should be possible to think about the ideas in geology, as well. How good is that bearing you measured on your compass? When should I count your answer as "wrong" - when it's half a degree from mine? One degree? Five degrees? Even if you weren't balancing on a cliff with shaking hands, would our measurements always be identical? When I talk about uncertainty in class, it's usually in that context.

Uncertainty in weather models is something I know less about. I know, from reading the discussion of the hurricane forecasters, that they look at the predictions of a number of different models (which presumably use different ways to weigh the effects of wind shear and warm Gulf water and other weather systems). I don't know where the 300 mile uncertainty comes from. (Is it the same kind of sense based on experience as I have about the correctness of my compass measurements? Or is there a more formal statistical way to derive the model's uncertainty?)

I hope that Gustav does not become a teachable moment. May the predictions be wrong, and may the storm weaken. And may everyone be prepared, in the event that it is correct.

5 comments:

John Fleck said...

I've always been struck by the fact that the hurricane forecasters sign their names. Think of all the anonymous paper products of the federal bureaucracy, but the hurricane forecasters, who grasp that lives are on the line, sign their names.

Kim said...

Reading the discussion is really interesting. There's a mix of very technical discussion and very human touches. (I remember one storm that defied predictions and left the forecasters noticeably frustrated.) The combination of the caps and the names is very different from most science communication products.

Elli said...

I have a personal interest in this one, since I'm driving to the Mississippi coast this Labor Day weekend (a friend is moving down there). And so I keep watching where the storm is going, glancing at what the evacuation recommendations are at, and wondering whether I'll actually manage to fly out of the Gulfport airport on Tuesday to teach on Wednesday. The airlines are offering to allow people to change their plans 7 days in either direction--something I may take advantage of once I actually make it down there.

The other thing I wonder about, though, is whether people 300 miles off from the main track are also preparing for the storm. Yes it might miss New Orleans, but Florida & Texas have also been the sight of some very strong hurricane damage.

Chuck said...

I think there's a site somewhere that shows all the model run tracks, if you're looking for a teaching moment...

Kim said...

It's the first day of class tomorrow, so I'm not going to be ready to do heavy teaching. If Gustav misses New Orleans (crosses fingers and toes), I might talk about uncertainty on Wednesday, though, and then it would be good to show the various model runs. (For Hanna, as well - the forecasters keep emphasizing how poorly the models handle the combination of factors controlling Hanna's track and intensity.)